General (Knowledge Base) Agent
A Swiss-army assistant: search, edit, digest, email, schedule.
Your personal knowledge manager: answers questions over your library, runs deep online literature searches, edits documents in place, sends daily digests, schedules recurring jobs, and runs data analysis on uploaded spreadsheets.
What it can do
- Semantic Q&A over your knowledge base with cited sources
- Deep online literature search across academic databases and the web
- In-place editing of Word/PDF documents with version history
- Full review-writing capability, callable in natural conversation
- Digest generation and email delivery
- Natural-language scheduled tasks (daily/weekly)
- Data analysis over Excel/CSV uploads with plotting
Best-fit tasks
- Rapid literature triage and Q&A
- Continuous paper tracking with automated weekly digests
- Editing and exporting papers, reports, contracts
- One-shot review writing grounded in the knowledge base
- Excel cleaning / visualisation / statistics
- Automating recurring chores via scheduled tasks and email
- Cross-library comparison across multiple knowledge bases
How to brief the General Agent
- First time: create or pick a knowledge base, upload your papers/reports/spreadsheets, optionally add tags.
- Q&A: just ask "what do papers in the KB say about X?" — answers come with quoted sources.
- Document editing: upload a Word/PDF and say "replace section 2 with a meta-analysis", edits happen in place with a new version.
- Literature push: "every Monday 8am search for new Agents papers, pick 5 and email to [email protected]" creates a scheduled job.
- Data analysis: upload an Excel file and say "compute quarterly sales and plot a trend line".
- One-shot email: "email the summary above to me" is enough.
Literature Review Agent
From paper collection to a finished review, end-to-end.
Built for researchers writing literature reviews, thesis review chapters, or standalone review papers. Covers the full pipeline: summarising papers, drafting the outline, writing the body, and iterating on revisions.
What it can do
- Batch-reads uploaded PDFs and extracts key points plus reference metadata
- Plans a review outline from your topic and paper pool, managing citation numbers automatically
- Writes the full review chapter by chapter, in Chinese or English, under your chosen citation style
- Supports both AI-driven section rewrites and surgical sentence-level edits, with automatic versioning
- Built-in academic literature search so you can add newer papers and have the outline re-numbered
Best-fit tasks
- Graduate students drafting the literature review chapter of a thesis
- Researchers preparing a submission-ready review article
- Getting up to speed in an unfamiliar field quickly
- Updating an existing review with the latest publications
- Producing Chinese/English bilingual versions of the same review
How to brief the Literature Review Agent
- Upload the PDFs you want reviewed to the session or knowledge base (10–20 per batch works best).
- In your first message state four things: (1) topic, (2) output language, (3) citation style (e.g. [1], APA, GB/T), (4) any length or section constraints.
- Once summaries finish, review the outline first — raise structural concerns now, not after the full draft.
- Iterate section-by-section or sentence-by-sentence: say “rewrite Section 3.2” for big changes, paste the original + target wording for small ones.
- When satisfied, ask it to “export as Word / PDF”.
Research Agent
All-in-one research assistant: data analysis, modeling, code execution & reproduction, AI imagery.
It unifies data analysis, statistical modeling, machine learning, data visualization, production-grade code engineering, paper code reproduction, and AI image generation into one agent. It works like an engineer: write a script, run it, read the results, and self-fix on errors. Upload data for analysis and high-quality charts, hand it a paper to reproduce its method code, or give it a complex multi-module project to build from scratch. It loads knowledge-base retrieval on demand when you need to consult your library.
What it can do
- File-based code execution: write, run, read results, self-fix on errors
- End-to-end analysis: cleaning → EDA → stats/modeling → visualization → report
- High-quality charts: unified design language, CJK fonts and grid details handled
- Paper reproduction: draft a plan first, then produce a modular runnable repo in stages
- Complex engineering: blueprint-first, file-by-file implementation, incremental verification
- AI imagery: concept diagrams, framework figures, cover images for reports
Best-fit tasks
- Analyze uploaded CSV/Excel and produce an illustrated report
- Regression, hypothesis testing, ML modeling and evaluation
- Reproduce a paper's method/model and run it to verify
- Build a multi-module data pipeline or modeling project from scratch
- Add professional data charts to back up conclusions
How to brief the Research Agent well
- Upload data/paper files and state your goal (what to analyze, reproduce, or build).
- Analysis: name the key variables and the question; let it clean → model → chart → conclude.
- Reproduction: give the paper; have it draft a plan first, then implement and run it.
- Engineering: describe project goal and scope; it plans the architecture then builds step by step.
- Large tasks proceed in steps; artifacts (scripts, charts, reports) land in the session dir.